A0414
Title: Conditional partial exchangeability: A probabilistic framework for multi-view clustering
Authors: Beatrice Franzolini - Bocconi University (Italy) [presenting]
Maria De Iorio - UCL (United Kingdom)
Abstract: Standard clustering techniques assume a common configuration for all features in a dataset. However, when dealing with multi-view or longitudinal data, clusters' shapes and definitions may need to vary across features to accurately describe structure and heterogeneity. This framework requires accounting for within-subject dependence across multiple features, even with different support spaces. Unfortunately, classical model-based clustering techniques fail to account for this dependence. A wide class of Bayesian clustering models designed is introduced to tackle this challenge. The core feature is conditional partial exchangeability, a novel probabilistic paradigm that handles the problem, induces dependence among partitions, ensures tractability, and could ultimately provide a probabilistic framework for the development and study of dependent random partitions of the same subjects.